Arlequin AI: A New AI Architecture for Critical Decision-Making

We are pleased to announce that OTB Ventures is co-leading Arlequin AI's €28 million Series A alongside redalpine, with participation from Bpifrance's Defence Innovation Fund, Vsquared Ventures, 10x Founders and Xavier Niel. The round is backed exclusively by European investors.

Bigger models are not always the answer. Better architectures can be.

That conviction is what drew us to Arlequin. Most of the capital in AI today is chasing the same architecture, betting that scale alone will solve every problem. Arlequin is taking a different route, building proprietary models based on topological neural networks (TNNs).

Why critical decisions need a different kind of AI

Founded in Paris in 2024 by Hugo Micheron and Antoine Jardin, Arlequin develops AI for environments where governments and large enterprises make consequential decisions from large, fragmented datasets, and where every result has to stay grounded in the underlying evidence.

This is a hard problem for the architectures that dominate AI today. Large language models have transformed language and content generation, but they were not designed to map the structures and dynamics hidden across millions of interconnected data points, or to show their work when the answer carries real-world consequences. In security, defence and investigative work, an estimate of the truth is not enough. What matters is the evidence behind each result.

A new architecture: topological neural networks

Topological neural networks learn not only from individual data points, but from how data is connected, including relationships that involve many elements at once. Arlequin is building the architecture to analyse increasingly large and intricate systems while keeping every result traceable to its source. The work is supported by collaborations with research teams at INRIA, CNRS and the Max Planck Institute, and at the Universities of Oxford, Cornell, Princeton and UC Santa Barbara.

The approach also runs against the prevailing cost curve of AI. The TNN architecture requires significantly less compute, reducing reliance on the energy, advanced semiconductors and computing infrastructure that increasingly constrain the development of large AI systems.

Built for high-stakes environments

Arlequin's platform analyses relationships across heterogeneous data, from documents and transactions to video and operational information. In a counterterrorism investigation, it can process millions of data points across multiple seized devices to surface connections between people, locations, communications and events. The same approach applies across security and defence, criminal, fraud and money-laundering investigations, information integrity, cybersecurity, and the safety and security of AI systems themselves, all domains where predictions alone are insufficient and the evidence behind each result is what counts.

A team at the intersection of research and product

We backed Hugo and Antoine because of an exceptional founder-market fit. Hugo Micheron, who holds a PhD in political science from the Ecole Normale Superieure (ENS), is recognized as one of Europe's leading experts on how jihadist networks structure themselves within European communities and online platforms. Antoine Jardin is a former CNRS research engineer in data science and human behaviour who was instrumental in setting up the Jean Zay supercomputer, one of the most powerful computing resources in France. Together they have assembled a team of around 50 people, including 35 engineers and 15 PhDs, researchers and data scientists.

What sets the company apart is that its fundamental research is the direct engine for a product already in use. Arlequin's technology is deployed with governments and large organizations across Western and Eastern Europe, and the company is active in four European countries. That combination of deep science and real commercial traction remains rare in Europe.

What's next

The funding will let Arlequin expand the international team developing and training its proprietary models and accelerate commercial deployment across Europe and worldwide. The company has opened offices in London and Berlin, and plans to establish an AI lab in Silicon Valley by the end of 2026.

Dual-use technologies are a core focus for OTB. Arlequin sits squarely within our thesis: backing European founders who build the critical infrastructure of the continent's technological future. AI sovereignty is not only about where models are built or where data is hosted. It is about control over the technologies that increasingly underpin our most critical decisions, and about AI systems that Europe can own, understand and trust. Arlequin is opening a path for Europe to compete by originating new fields of AI rather than reproducing existing ones.

Full press release: https://arlq.ai/news/arlequin-ai-series-a-topological-neural-networks